Up to four models
DeepSeek-V3 (Mar 2025) vs DeepSeek-V3
On the index (GPT-4 = 100), DeepSeek-V3 (Mar 2025) scores 124 and DeepSeek-V3 119. Side by side on the same third-party tests, with price and context. The link in your address bar shares this exact comparison.
124108–140
WeightsOpen
ReleasedMar 2025
Price in / out$0.29 / $1.14
Context164K
Core tests taken2 of 9
119103–135
WeightsOpen
ReleasedDec 2024
Price in / out$0.26 / $1.03
Context164K
Core tests taken2 of 9
DeepSeek-V3 (Mar 2025) leads DeepSeek-V3 by 5 points, ahead on 2 of 2 shared core tests, at a similar price.
Test by testthird-party results
Every result as a table
| Test | DeepSeek-V3 (Mar 2025) | DeepSeek-V3 |
|---|---|---|
| Reasoning & knowledge | ||
| GPQA Diamond (Epoch AI) | 67.6% | 56.5% |
| SimpleBench | 27.2% | 18.9% |
| Maths | ||
| AIME-style maths (OTIS mock) | 37.8% | 15.8% |
| MATH (level 5) | 75.5% | 64.8% |
| Coding | ||
| Aider Polyglot | 55.1% | 48.4% |
Task length · METR
30 sec
28 sec
Tests at least two of these models have taken. Bold rows are the core tests.